Building a DIY Adsorption Chiller — Working Pair, Vacuum, Bed, Evaporator

This is the build layer under Adsorption Cooling and its performance/bed references. The wiki is rich on the physics, materials, and sizing; this reference collects what’s known about actually fabricating one — the working-pair decision, the five subsystems, and the vacuum/leak-tightness problem that stalls most amateur attempts. It is the adsorption-side sibling of DIY PCM Encapsulation.

Reality check. Fully documented amateur adsorption builds are rare — the maker web has discussion and incomplete attempts but almost no start-to-finish build with a sealed/evacuated vessel and a measured result. The best reproducible references are student bench prototypes and academic solar ice-makers. The recurring stall point is sealing and holding a high vacuum without proper equipment. Start with the safe zeolite/silica–water route before ever attempting a methanol or ammonia system.

⚠️ Safety first. Methanol is toxic (ingestion/skin/inhalation → blindness, CNS/organ damage) and flammable; its sub-atmospheric loop must be leak-tight both to keep air out and methanol in. Ammonia is toxic and corrosive and runs above atmospheric pressure (pressure-rated components only). Deep vacuum vessels can implode — use rated vessels and eye protection. If you are not equipped to handle toxic/flammable refrigerants, build the water system, which uses none.

Step 0 — pick the working pair (it decides safety, vessel, and performance)

Pair	Vessel	Cold reachable	Performance	Safety	DIY verdict
Zeolite/silica + water	deep-vacuum (~1–23 mbar)	~10 °C (0 °C freeze floor)	bench COP ~0.4, evap 9.8 °C	safe — no toxic refrigerant	Start here. Hard part is only the vacuum
Activated carbon + methanol	moderate vacuum (~3 kPa evap)	sub-zero, makes ice	solar COP ~0.12, ~6 kg ice/m²/day	toxic + flammable	Canonical solar ice-maker; hazard-managed builds only
Activated carbon + ammonia	pressure tubing (above atmos)	sub-zero, makes ice	solar COP ~0.05, ~3 kg ice/m²/day	toxic + corrosive	Easy leak detection, no vacuum — but pressure + toxicity

Water has the best latent heat and is the only non-hazardous option, but it can’t go below 0 °C and needs the deepest vacuum. Methanol makes ice at a moderate vacuum. Ammonia dodges the vacuum entirely (it runs above atmosphere, so leaks hiss out rather than suck air in) at the cost of pressure-rated plumbing and toxicity. Adsorbent:refrigerant mass ratio is roughly 4:1 (e.g. 42 kg carbon : 10 kg methanol).

The five subsystems

1 · The vessel. Water/methanol → a leak-tight vacuum vessel; ammonia → pressure-rated tubing. DIY vacuum vessels are welded stainless pipe with ISO/KF flanges and O-ring seals — threaded + PTFE-tape joints leak under real vacuum. The simplest topologies: two valved cylinders (bed + water, Chachad & Bhadane), or a single glass tube with the bed at one end and a combined evaporator/condenser at the other (Khattab, valveless). Anjorin & Bello’s field build used one single valve at the condenser for both evacuation and charging — a nice parts-count reduction.

2 · The adsorbent bed. Buy silica gel (0.5–1.5 mm) for water, coconut-shell activated carbon for methanol/ammonia, or zeolite 13X/4A for higher uptake. Activation is mandatory before charging — bake out adsorbed water: silica gel at 60–85 °C (domestic oven); zeolite 13X at ~150 °C (but don’t over-bake — 13X loses 31/51/68% capacity after 5000 cycles at 150/200/250 °C). For silica/water, degas 150 °C under vacuum 8–12 h. Pack granules against corrugated fins; the coat-first-then-pack trick removes metal-to-grain contact resistance. Cheapest conductivity boost: pack the sorbent into open-cell aluminium foam bonded to the tube (+50% adsorption rate). Keep the bed ~5 mm thick with ~5 mm vapor channels — both heat- and mass-transfer resistance scale with thickness². Budget ~20–26 kg sorbent per few kW.

3 · The evaporator (where the cold is made). For water under vacuum, the killer detail is the hydrostatic penalty: a deep pool raises local saturation temperature (20 cm depth → +3.2 kPa → boils at 25 °C, no cooling), so keep it shallow. Thimmaiah/Bahrami’s copyable flooded design: a bundle of 1/2” finned copper tubes (26 fpi) in a shallow pool inside a sealed box, chilled water pumped through the tube bores, refrigerant charged to H=0.8* (80% of tube diameter). That gave 135–440 W over a 5–20 °C chilled inlet. Best pumpless option is capillary-assisted (fine 40-fpi fins wick from a shallow pool — near falling-film with no pump). Or skip the loop entirely and make the evaporator itself an ice tray.

4 · The condenser + receiver. An air-cooled finned coil (natural convection for small units; water-cooled gives a lower condensing temperature and better COP) rejects the desorbed vapor, draining to a receiver that holds the charge and feeds the evaporator by gravity/capillary (valveless) or an expansion valve.

5 · The heat source. 60–95 °C drives silica/water; a flat-plate solar collector whose black plate is the adsorber reaches ~90–100 °C by day (the classic intermittent solar cycle). Zeolite needs ~150–200 °C to regenerate.

Vacuum & leak-tightness — the make-or-break

Water refrigerant boils near 6 mmHg (~0.8 kPa) at chilling temperatures, so the system runs under deep vacuum and any tiny leak steadily accumulates air that blankets the heat-exchanger surfaces and kills capacity (~8–14% loss). Procedure:

Pump: an HVAC rotary-vane pump. For deep-evacuation practice (HVAC habit) a two-stage pull to ≤500 µm is ideal — but for this water demonstrator a cheaper pump is genuinely enough (see below): the operating point is only ~10–23 mbar, so anything reaching a few hundred microns has 10×+ margin. Gas ballast (or an upstream cold trap/desiccant) matters more than the extra stage, because water vapor is what wrecks the oil.
Nitrogen sweep 2–3× before evacuating to displace moist air; then run the pump 30–60 min (hours if saturated) to boil off residual water vapor.
Decay / rate-of-rise test: isolate the pump, watch 10 min — pass = <200 µm rise, fail = >500 µm. Mount the gauge at the vessel, not on the manifold.
Read real vs virtual leaks on a linear scale: a true air leak rises fast then keeps climbing at a steady slope; outgassing/moisture (trapped in O-rings, threads, the water itself) rises then tails off flat. Don’t chase phantom leaks — this is the single most important skill.

Choosing a low-cost pump (for the water demonstrator)

The water side only has to reach its saturation pressure at the target temperature — ~23 mbar (17.5 Torr) for 20 °C, ~12 mbar for 10 °C, ~6 mbar for 0 °C — so this is rough vacuum, and at a few litres pump-down takes seconds. Ultimate pressure and surviving water vapor are the real constraints, not speed.

Recommended — budget/used HVAC rotary-vane pump with gas ballast (~$60–100 used, ~$120–130 new; e.g. VEVOR VP245, Harbor Freight ICON 58141, used JB Eliminator). Rated 15–150 µm — 100×+ deeper than needed, and single-stage suffices (a 300-µm stall under vapor load is still far below the ~10,000-µm operating point). This is the right buy.
Cheapest that works — two salvaged fridge/AC compressors in series (free–$30). One alone bottoms at ~20 Torr (marginal); two in series clear it. Keep the oil topped up.
Most water-appropriate — cold-water aspirator (~$15–40, oil-free, contamination-free). Capped by water vapor pressure (~24 Torr at 25 °C), so it needs genuinely cold supply water to reach ~15–20 Torr; wastes water.
Don’t bother: Mityvac/hand pumps (~125 Torr), diaphragm pumps (50–100 Torr), single-stage venturis (60+ Torr) — their inHg specs look close but are 3–7× short in absolute terms.

Full comparison: vacuum-pump options note.

Measuring the vacuum (which gauge)

The rig’s two measurement jobs straddle exactly where thermal (“micron”) gauges fail:

The purge (Job 1) — reaching a few hundred microns to strip air — is where a Pirani/thermocouple micron gauge is accurate (±10% below 10 Torr).
The operating point (Job 2) — ~10–23 mbar of nearly pure water vapor — is the worst place for one: Pirani accuracy collapses to ±50% above 10 Torr, and a nitrogen-calibrated thermal gauge over-reads water vapor ~40–60%.

The trick: read operating pressure off the water thermocouple. Near 10–23 mbar the saturation curve is ~1 mbar/°C, so the evaporator temperature the experiment already logs is your operating-pressure gauge (±1 °C ≈ ±1 mbar). You don’t need a second gauge for the operating point.

Recommended — budget micron gauge for the purge only. Pick one rated to ~19,000–25,000 µm (Elitech VG-760/VGW-mini ~$70–90; NAVAC NMV1S / Fieldpiece MG44). Range trap: Fieldpiece SVG3 (9,999 µm) and JB DV-41 / old Supco VG64 (12,000 µm) go off-scale below the ~17,500 µm operating point.
Optional gas-independent operating readout — MS5803-01BA (~$10–30 + Arduino): 10–1300 mbar, 0.012 mbar resolution, reads true water-vapor pressure. Or a sealed mercury U-tube (23 mbar = 17.3 mm Hg) as a cheap absolute reference.
Don’t bother: compound Bourdon gauge (whole 0–25 Torr in the last mark), MPX5100 (±2.5 kPa error > signal), BMP280/388 (30 kPa floor).

Full comparison: vacuum-gauge options note.

Performance to expect

Zeolite/silica–water bench unit: evaporator down to ~9.8 °C, COP ~0.4 (Chachad & Bhadane). A cooling demonstrator, not an ice-maker.
Solar carbon/methanol ice-maker: ~6 kg ice per m² of collector per day at ~20 MJ/m²/day, solar COP ~0.12 (Pons & Guilleminot); replications report COP 0.13–0.16 and 4–9 kg ice/m². ~130 kg carbon over 6 m² → 30–35 kg ice/day.
Solar carbon/ammonia: ~3 kg ice/m²/day, COP ~0.05 (Critoph).
Powered silica-gel/water chiller (the 601-scale target, not a first build): COP ~0.45, SCP ~176 W/kg at 80/30/14 °C — see Performance & Numbers.

Two recommended build paths

Safe demonstrator → zeolite/silica + water. Two valved cylinders (or one vessel with a shallow finned-copper evaporator), ~0.6 kg zeolite : 500 mL water at bench scale. Bake the sorbent, evacuate the water side to ~23 mbar, sequence the valves so the sorbent vapor-pumps the water. Expect chilled water ~10 °C, no ice, no toxic refrigerant — the right first build to learn the vacuum craft.
Solar ice-maker → activated carbon + methanol (only with methanol-handling safety in place). Black-plate flat-plate collector as the adsorber (~40 kg carbon per 1–1.5 m²), single valve for evacuate+charge, air-cooled finned condenser, submerged-plate evaporator as the ice tray. Day desorbs, night freezes. Expect ~5–9 kg ice/day per m². Manage flammability/toxicity throughout.

Avoid, for a first build: a continuous two-bed powered chiller (needs synchronized valving + heat/mass recovery) and any ammonia system (pressure + toxicity).